Can time-to-detection models with fewer survey replicates provide a robust alternative to traditional site-occupancy models?

Occupancy models are widely used in ecology because they explicitly separate the observation and state processes and hence account for imperfect species detection. Traditional occupancy models that record detection/non-detection (DND) of a species typically rely on either spatial or temporal  survey replication to estimate model parameters. Recording the time until a species is first encountered after starting a survey is often possible with little extra effort and such time-to-detection (TTD) surveys may be more efficient than pure DND surveys. Using continuous time data, TTD occupancy models can in theory estimate occupancy and detection parameters using a single TTD survey. These models therefore have the potential to drastically reduce the logistical effort and costs associated with traditional occupancy survey designs. However, the robustness and general applicability of TTD models has not been widely addressed and their effectiveness in different study systems remains unknown.
We use simulations and bird data of 63 species from a field study in the Karoo region of South Africa to explicitly compare estimates of occupancy, detection and species richness between DND and TTD models under various levels of survey replication and for species with different occupancy and detection characteristics.
Simulations revealed that for inconspicuous species (low detection probability) single survey TTD models can perform better or equally as well as DND models with a higher number of replicates. This effect was attenuated in widespread species (high occupancy probability). The benefits of TTD models were more pronounced at low survey replicates and performance of the two methods converged quickly as the number of survey replicates increased. The difference in model performance related to precision around estimates while the bias in parameter means was fairly low. However, results from the field data showed that a single TTD survey was not adequate to reliably estimate occupancy, detection and species richness; especially in rare and inconspicuous species. Increasing the number of TTD surveys to two replicates improved the models substantially.
Our results demonstrate the general utility of TTD surveys depends on the characteristics of the species considered in the study. A single TTD survey may be sufficient in some study designs but is unlikely to be sufficient in most multi-species field scenarios where communities are made up of species that have a wide range of detection and occupancy probabilities. TTD surveys do provide benefits however in that data can be used to construct detection curves which can be used to guide survey effort in the design of future studies.

占用模型（Occupancy models）在生态学研究中应用广泛，因其可明确区分观测过程与状态过程，从而能够校正物种检测不完全的问题。传统的检测/非检测（DND）占用模型，通常需借助空间或时间层面的调查重复来估算模型参数。在启动调查后记录物种首次被检测到的耗时，通常仅需额外投入少量工作，且这类检测耗时（TTD）调查相较纯DND调查效率更高。理论上，依托连续时间数据，TTD占用模型仅需单次TTD调查即可估算占用率与检测参数。因此，此类模型有望大幅削减传统占用调查设计所需的后勤成本与工作量。然而，TTD模型的稳健性与普适性尚未得到广泛探讨，其在不同研究系统中的有效性仍未明确。
本研究通过模拟实验，结合南非卡鲁（Karoo）地区野外研究中63种鸟类的观测数据，在不同调查重复水平下，针对具备各异占用率与检测特征的物种，明确比较了DND模型与TTD模型在占用率、检测率及物种丰富度估算方面的表现。
模拟结果显示，对于隐蔽性物种（低检测概率），单次调查的TTD模型，其表现可优于或等同于高重复次数的DND模型；而在广布物种（高占用率）中，这一效应会被弱化。TTD模型的优势在低调查重复水平下更为突出，且随着调查重复次数增加，两种方法的性能会快速趋同。两种模型的性能差异主要体现在估算值的精度上，而参数均值的偏差相对较低。不过，野外数据的结果表明，单次TTD调查不足以可靠估算占用率、检测率及物种丰富度，尤其是针对稀有且隐蔽的物种。将TTD调查次数提升至两次后，模型性能得到显著改善。
本研究结果表明，TTD调查的通用实用性取决于研究中所涉及物种的特征。在部分研究设计中，单次TTD调查或可满足需求，但在多数多物种野外场景中，当群落由检测率与占用率跨度极大的物种组成时，单次TTD调查往往难以达到要求。不过，TTD调查确实具备独特优势：其采集的数据可用于构建检测曲线，进而为未来研究设计中的调查工作量规划提供指导。